Place cells are neurons that respond to a particular location in the physical world. For example, as a rat runs around a maze, some place cells will become active when the rat reaches one corner. When the rat moves on towards a different corner, other place cells activate instead. The real-time activity of these place cells helps the rat to work out where it is in the maze. This activity also contains information about what the rat is doing.

In addition to their real-time activity, place cells also help previous events to be ‘replayed’ mentally, which is important for making decisions. Previous studies have shown that when a rat pauses during a task, place cell replays allow it to mentally map out the route it needs to take. However, it is less clear whether these replays also provide information about what the rat needs to do.

Takahashi gave rats a number of tasks to perform inside a figure of eight maze. In one of these tasks, the rat had to assess which one of two lights was lit up, and run towards it. In the other two tasks, the rat had to remember the direction it took on the previous occasion, and go in the opposite direction. During these tasks, the rat would occasionally pause to replay information about the task.

Takahashi recorded what the rats' place cells were doing during these pauses, and found that the place cell replays contained information about both the path the rat needed to take (‘where and when’ information), and which task it needed to carry out (‘what’ information). This suggests that replays are important for the ability to recall information about specific events, which is known as episodic memory. Takahashi's results may therefore also help us to learn more about this “mental time travel” and human conditions that damage episodic memory, such as Alzheimer's disease.

One of the most important aspects of episodic memory is the mental retrieval of personal recollections of what happened “where” and “when”, referred to as “mental time travel” (Tulving, 1983). Episodic memory retrieval is important not only for remembering recent events but also for imagining future ones (Hassabis et al., 2007). The hippocampus is critically involved in episodic memory retrieval, and within this structure reside place cells—principal cells exhibiting place-specific firing patterns in particular locations. The presence of such cells has led to the hypothesis that the hippocampus is the locus of the “where” element of episodic memory (O'Keefe and Dostrovsky, 1971). Following on from this seminal finding, more recent studies have demonstrated that hippocampal place cell activity simultaneously represents a broad range of aspects of memory, including odors and their match/non-match status in a particular location, called the “place field”, as well as elapsed time (Otto and Eichenbaum, 1992; Sakurai, 1996; Wood et al., 1999; Pastalkova et al., 2008; MacDonald et al., 2011). In the place field, both firing rate and location can change independently and are affected by external factors, such as spatial environment and task demand, and internal factors such as prospective and retrospective memory (Markus et al., 1995; Frank et al., 2000; Wood et al., 2000; Anderson and Jeffery, 2003; Ferbinteanu and Shapiro, 2003; Smith and Mizumori, 2006). During maze running, place cell activity sequences in rodents convey information on both familiar paths and accompanying subtasks (Allen et al., 2012; Takahashi, 2013).

Within sharp wave/ripples (SWRs) in local field potentials (LFPs) during slow-wave sleep or periods of awake immobility, internally generated place cell activity sequences are often reactivated in a temporally compressed manner (Wilson and McNaughton, 1994; Foster and Wilson, 2006; O'Neill et al., 2006; Diba and Buzsáki, 2007; Karlsson and Frank, 2009; Gupta et al., 2010). During brief periods of immobility, pivotal paths to a remembered goal can be predicted, regardless of the location in which the SWRs have previously occurred (Pfeiffer and Foster, 2013). Such “replay” is thus considered a neuronal substrate for retrieving memories on familiar paths (Carr et al., 2011). However, previous studies investigating the content of replays have primarily focused on the geography and timeline of the animals' running path as experienced in the environment. It is therefore unclear whether the replay conveys information on events (i.e., “what” information) occurring along the reactivated path.

I recently recorded the ensemble activity of place cells in the hippocampal CA1 of rats made to continuously navigate four journeys in a figure-eight maze under three intermittently switching subtasks within a single session: visually guided discrimination (VD), non-delayed spatial alternation (NA), and delayed spatial alternation (DA) (Takahashi, 2013). This task design allowed for comparison of hippocampal neuronal activity between and within different journeys and subtasks. I found that while the animal is running in the maze, differences between individual journeys and between subtasks in the maze are independently encoded in the firing locations and rates of place cells, respectively. Furthermore, all trial types (i.e., journey type and subtask) experienced in the maze could be decoded from the ensemble activity of place cells. Since the rats ran along similar spatial paths while performing different subtasks throughout this task, differences between subtasks can be interpreted as non-spatial “what” information. Provided that the mechanisms are preserved in temporally compressed replays that occur during brief periods of immobility, the subtask as “what” information occurring along the path in the maze might be examined in the replay. Building on this previous study (Takahashi, 2013), I accordingly investigated whether “what” information is contained in the replay of place cell activity sequences during the brief immobility periods that occur while the animal is engaging in the task.

According to the subtask switching protocol, four rats were trained for about one hour in a specific sequence—VD (20 laps), NA (20 laps), VD (10 laps), DA (20 laps), VD (20 laps), NA (20 laps), VD (10 laps), and DA (20 laps)—and were rewarded each time they arrived at the reward zone at the correct return rail (see ‘Materials and methods’). In the VD subtask, a cue light on the left or right side was illuminated to indicate which direction to turn at a decision point (Figure 1A) to receive a reward. In both the NA and DA subtasks, cue lights on both sides were illuminated and the rats had to alternate between left and right at the decision point (Figure 1B). The difference between the NA and DA subtasks was that in the DA subtask, a barrier appeared in the middle of the central stem for five seconds to impose a delay period. The subtask to be rewarded (i.e., VD, NA, or DA) could therefore only be distinguished at the decision point, based on visual cues (Takahashi, 2013).

Figure 1 with 1 supplement see all

Download asset Open asset

Rat behavior during brief periods of immobility

Before examining the replays, I investigated the rats’ behavior during the brief periods of immobility, when the replays occurred and the animals' behavior suggested that they were actively gathering information (Benjamini et al., 2011). Specifically, I assessed the directions in which the rats' heads were pointing at these times. In correct trials, the rats' heads pointed towards a goal in the opposite direction from the respective start zone (Watson–Williams test, P: 1.1 × 10⁻¹⁶, n = 588 (left), 1305 (right); Rayleigh test; P: left, 1.1 × 10⁻⁵⁶, right, 4.9 × 10⁻³²⁴) and stem (Watson–Williams test, P: 1.1 × 10⁻¹⁶, n = 257 (left), 395 (right); Rayleigh test; P: left, 1.7 × 10⁻²³⁵, right, 5.2 × 10⁻¹²⁵) (Figure 2). This suggests that the rats sampled a memory-guided goal in preparation for switching to either the NA or the DA subtask. In this experiment, head direction is therefore considered a behavioral sign indicating expected future choice. In erroneous trials, however, head direction was ambiguous only in the maze stem (Figure 2C,D; Watson–Williams test, (2C) P: 4.4 × 10⁻¹⁶, n = 29 (left), 30 (right), (2D) P: 0.29, n = 26 (left), 10 (right); Rayleigh test, P: (2C) left, 3.1 × 10⁻⁵, right, 8.2 × 10⁻¹⁸, (2D) left, 2.0 × 10⁻¹⁴, right, 5.1 × 10⁻⁷). Considering that the barrier in the maze stem would have been an indication to the rats of the ongoing DA subtask, this suggests that the spatial memory demand in preparation for NA/DA subtasks decreased within the maze stem. In addition, prior to making erroneous choices, the rats did not make a decision at the behavioral level within the maze stem.

Figure 2

Download asset Open asset

Running path in the replay

I examined path representation in the temporally compressed replay of place cell activity sequences during the brief periods of immobility in the awake behavioral state. Some reactivations of place cell activity sequences during SWRs were related to the rats' upcoming path on the task. For instance, consider the spike raster plots of nine replays that occurred in the start zone, central stem, or junction (Figure 3). To arrange the sequences in terms of the cells' place fields along the journey, I ordered the cells by the latency of their peak firing rates during the corresponding trial type. The cell number decreases from the start zone to the goal. Sequential reactivation of the place cell activity during the replays shown in the raster plots began where the rat paused (blue circle in the trial-type illustration) and moved forward toward a goal.

Figure 3 with 3 supplements see all

Download asset Open asset

To test whether the replays consistently related to upcoming behavior, I used the Bayesian decoder at a 10-fold compressed scale (Davidson et al., 2009; Pfeiffer and Foster, 2013) in the replay analyses. To avoid confusing replay and sequential activation during movement-related phase precession, population bursts of spiking activity during periods of awake immobility, when rats were moving less than 2 cm/s, were identified as candidate replays (Figure 3—figure supplement 1A). Of a total of 2820 candidates, 552 were identified as statistically significant replays with a continuous path, whose length and duration satisfied several criteria (Davidson et al., 2009; Pfeiffer and Foster, 2013) (see ‘Materials and methods’; Figure 3—figure supplement 1; Table 2). LFPs during the path replays were largely coincident with the SWRs (99.8%; Table 2), and their theta power (4–12 Hz) was significantly smaller than during running (Figure 3—figure supplement 2). This confirms that the path replays identified here were similar to those reported previously (Foster and Wilson, 2006; Diba and Buzsáki, 2007; Davidson et al., 2009; Pfeiffer and Foster, 2013) and did not include low-speed replay events associated with theta oscillation (Johnson and Redish, 2007; Jezek et al., 2011).

Table 2

	Correct trials
	Candidate-start		Candidate-stem
Number	1893			810
Replay type	Path-start	Episodic-start	Path-stem	Episodic-center	Episodic-junction
Number	345	71	190	17	17
Percent confirmed	18.2%	20.6%	23.5%	21.5%	15.3%
Percent SWR coincident	100%	100%	99.5%	98.7%	100%

	Erroneous trials
	Candidate-start		Candidate-stem
Number	59		58
Replay type	Path-start	Episodic-start	Path-stem	Episodic-center	Episodic-junction
Number	9	1	8	0	2
Percent confirmed	15.3%	12.5%	13.8%	0%	25.0%
Percent SWR coincident	100%	100%	100%	–	100%

1. SWR, sharp wave/ripple.

To examine the spatial specificity of path representation in the replays, I analyzed the posterior probability of locations decoded from path replays by the Bayesian decoder. The middle portions of Figure 3 show nine representative decoded paths from the spiking activity shown in the raster plots during candidate replays. As the spike raster plots suggest, the decoded paths also began where the rats paused and moved forward toward a goal in the corresponding trial type. In addition, despite the fact that the actual goal in the VD_RR trial was on the right side, the decoded path ended on the left side (green arrow), suggesting that the replay represents an upcoming path to a memory-guided goal (i.e., opposite to the previous goal). Across all replays, the posterior probability of decoded locations was spatially concentrated around the vicinity of the goal (Figure 4A). To statistically compare the strength of path representation in the replays between selected regions, path representation strength was defined as the sum total of the posterior probability. Path representation in the vicinity of both goals was significantly greater than in the remainder of the maze area (Figure 4B; Wilcoxon rank-sum test, p = 3.3 × 10⁻¹⁰⁶, n = 552). As awake replay often begins at the animal's current location (Foster and Wilson, 2006; Diba and Buzsáki, 2007) and is enhanced by rewarding outcomes (Singer and Frank, 2009), the replay may be initiated at the goal. Although reward expectation increases occupancy time and place-field representation around the goals, such initiation bias was not correlated with either occupancy time (Figure 4—figure supplement 1A,C) or spatial distribution of place-field representation (Figure 4—figure supplement 1B,D). This suggests that the path replays were not a simple consequence of either reward expectation or place-specific firings in the location where they occurred (Gupta et al., 2010; Carr et al., 2011). Rather, they may represent paths on a memory-guided journey towards an upcoming goal (Pfeiffer and Foster, 2013), or on the immediately preceding journey (O'Neill et al., 2006; Karlsson and Frank, 2009).

Figure 4 with 1 supplement see all

Download asset Open asset

Replay represents a path to a memory-guided goal

To test this hypothesis, I examined the representation of future and past journeys in the replays. The path replays predominantly occurred in the start zone and in the maze stem (Figure 2E,F). Unlike the start zones, the maze stem was common to all journeys. There were therefore two types of path replays: path-start and path-stem. I centered the posterior probability of decoded locations on the start position and rotated it according to the direction of the memory-guided goal. As expected from the initiation bias, the rats' current location was strongly represented in the posterior probability of decoded locations in path-start replays (Figure 5A). To statistically test the difference between future and past journeys, I defined future and past regions as those from the start position to memory-guided and previous goals, respectively. It is worth noting that the memory-guided goal was always opposite to the previous goal even in visually guided trials (VD_LL, VD_RR), because the task design did not allow the rats to identify the subtask until they reached the junction (see ‘Materials and methods’). Path representation (i.e., the sum total of the posterior probabilities of decoded locations) of future regions was significantly greater than that of past regions in the correct spatial alternation trials (Figure 5B; Wilcoxon signed-rank test, p = 0.0037, n = 345). Similarly, future regions were strongly and statistically significantly represented in the correct visually guided trials (Figure 5C,D; p = 0.0078, n = 52). In erroneous trials, path representation of future regions was slightly greater than that of past regions (Figure 5E), but the difference was not significant (Figure 5F; p = 0.73, n = 9). However, due to the physically different start zones of different journeys, these characteristics may be affected by the place-specificity of the locations in which the replays occurred.

Figure 5 with 1 supplement see all

Download asset Open asset

To compare the representation of past and future journeys in the path-stem replays, I centered the posterior probability of decoded locations on the rat's physical location at the time of occurrence, rotated it according to the direction of the memory-guided goal, and scaled it according to the distance to the memory-guided goal. Path representation was similar to that in the replays that occurred in the start zones (Figure 5G,I; Figure 5H: p = 9.8 × 10⁻⁸, n = 190; Figure 5J: p = 0.034, n = 38). Surprisingly, however, path representation in path-stem replays on journeys toward an erroneous goal (Figure 5K) was significantly different from that on journeys toward a correct goal (Figure 5L; p = 0.0078, n = 8), despite the fact that the rats sampled both goal sides evenly. Similar to the findings of a previous study (Pfeiffer and Foster, 2013), this suggests that in the maze stem, where spatial memory retrieval is a prerequisite for predicting a future goal, path representation in the replay is directly linked to the animal's future path.

The exact path encoded in the replay may be decoded from the locations showing the highest a posteriori probability. I therefore reanalyzed the path replays using point estimates of locations based on a posteriori probability. Similar to the posterior probability analyses, the decoded locations in the replays tended to represent an upcoming journey to a memory-guided goal (Figure 5—figure supplement 1). To corroborate the results from the Bayesian decoder, I furthermore applied a simple spatial reconstruction algorithm, based on place fields during running, to the path replays. To estimate path representation in the replay, the place maps of the maze in neurons participating in the replay were simply summed up in response to the number of spikes (see ‘Materials and methods’). Similarly to the results obtained using the Bayesian decoder, the path representation estimated using the simple spatial reconstruction algorithm also largely predicted an upcoming path to a memory-guided goal (Figure 6). A previous study (Pfeiffer and Foster, 2013) reported that the awake replays represent an upcoming path to a memory-guided goal, but did not examine erroneous behaviors. My findings strongly support this view, with the additional finding that the upcoming path encoded in the awake replay can be linked to the animal's future actions.

Figure 6

Download asset Open asset

Replay is enhanced by spatial working memory demand

Although hippocampal place cell activity per se represents non-spatial information as well as spatial information (Wood et al., 1999), the question of whether replays include non-spatial information remains to be addressed. I next examined whether place cell activity during replays can convey information not only on the path but also on the subtask occurring along it, similarly to place cell activity occurring during running.

I observed that the firing rate of place cells participating in replays during different subtasks was dramatically different between subtasks even if the replay depicted a similar decoded path. For instance, consider the firing rate histograms for participating neurons in replays occurring in the start zone, central stem, or junction (Figure 3). As expected from the timing of temporally compressed firings across place cells, the maximum firing rates of place cells during replays (∼80 Hz) were about 10 times faster than typical maximum firing rates during running (∼8 Hz). To estimate the exact temporal compression rate of place cell firing during replays, I divided the average interspike interval during candidate replays by that during the entire rest of the session. The median of the estimated compression rate was 9.0 (Figure 3—figure supplement 3), suggesting that similarly to the timing of firing, place cell firing rates were also temporally compressed during replays. During replays occurring at the junction, the raster plots of spiking activity in DA_RL and VD_RL trials recorded from rat II showed a similar activity sequence (Figure 3C, rightmost, arrows). The replays also depicted a similar decoded path. However, the firing rates during DA_RL trials were substantially greater than during VD_RL trials, suggesting that the firing rates of neurons participating in the replay may contain information on the differences between subtasks.

To test whether subtask information is encoded in the firing rates in the replay, I predicted trial types based on the firing rates of a group of cells recorded during path replays, using the prediction method in conjunction with the decoded locations (see ‘Materials and methods’; Figure 3—figure supplement 1). It was necessary to ensure that the replays used predicted a trial type with a probability greater than that dictated by chance. Accordingly, of the total 552 path replays, 108 were classified as episodic replays that satisfied criteria concerning the most frequently predicted trial type and its predicted probability sum (see ‘Materials and methods’; Table 2). As expected, only the replays that occurred in the junction (Figure 3C) appeared to predict the actual trial type correctly.

To examine this further, I investigated preference for subtask (i.e., VD, NA, DA subtasks). I examined whether there was any preference in the replays for particular rewarded subtasks. As expected, based on a previous study suggesting that awake SWRs support spatial working memory and not reference memory (Jadhav et al., 2012), the occurrence rate of overall episodic replays across all actual subtasks to be rewarded was significantly biased toward the DA subtask, which required spatial working memory (Figure 7A; Chi-square goodness-of-fit test, post hoc multiple binomial tests corrected by the Benjamini and Hochberg procedure, VD vs DA: p = 8.6 × 10⁻⁵, NA vs DA: p = 9.9 × 10⁻⁴). The null hypothesis tested was that the occurrence rates of predicted subtasks were equal to the rates expected based on the ratio of subtasks experienced throughout the entire session (60 laps VD: 40 laps NA: 40 laps DA).

Figure 7 with 3 supplements see all

Download asset Open asset

Does this bias depend on where the replay occurred? To address this question, episodic replays were divided into two types according to the occurrence location of the replay: episodic-start and episodic-stem replays. This was done because the head direction analyses suggested that spatial memory demand at the behavioral level was lower within the maze stem. The partitioning showed that the subtask bias occurred in episodic-start replays (Figure 7B; VD vs DA: p = 3.7 × 10⁻⁹; NA vs DA: p = 5.1 × 10⁻⁷), but not in episodic-stem replays (Figure 7C). Similarly, in the path-start replays, all candidates and SWR events showed such a bias (Figure 7—figure supplement 1A–C). This bias may be accounted for by occupancy time while the rats paused. However, since the rats specifically spent more time pausing in the DA subtask in both the start zones and the maze stem (Figure 7—figure supplement 1D,E), the bias cannot be explained by the time spent pausing. Combined with behavioral signs of decreased spatial memory demand within the maze stem, these results suggest that the occurrence rate of replays is enhanced by upcoming spatial working memory demand. In contrast, the subtasks predicted from the episodic replays did not show such a bias, and instead the prediction rates reflected the ratio of subtasks experienced throughout the entire session (Figure 7D). This suggests that the enhancement is not specifically due to the reactivated DA subtask, but instead due to the reactivation of all previously experienced subtasks.

Replay can predict the actual trial type to be rewarded

Examination of trial type, composed of journey type and subtask, should provide further understanding of subtask preference. Even if the replay can represent an upcoming path, disambiguation of the subtasks is still required for identifying the actual trial type. I therefore investigated the relationship between the actual trial type to be rewarded and trial-type representation in the episodic replays. Since the path representation suggested that the replay encodes an upcoming path to a memory-guided goal, I only examined the memory-guided trial types (i.e., VD_LR, VD_RL, NA_LR, NA_RL, DA_LR, and DA_RL). Trial-type representation in the episodic replays did not match the actual trial type in the start zones (Figure 7E; Choen's κ = 0.017; z test, p = 0.83, n = 71), but did match it in the stem (Figure 7F; Choen's κ = 0.31; p = 8.4 × 10⁻³, n = 34). This discrepancy may be the result of the extent of spatial working memory demand at the location where the replay occurs.

To further examine trial-type preference, I divided the episodic-stem replays into two types: episodic-center and episodic-junction replays. Trial-type representation matched actual trial type moderately well in replays at the junction (i.e., the decision point) where the rats could evidently identify the actual trial type based on the visual cue (Figure 7H; Choen's κ = 0.58; p = 3.7 × 10⁻⁴, n = 17), but not at the center of the stem (Figure 7G; Choen's κ = 0.022; p = 0.93, n = 17). To corroborate these results, the Bayesian decoder in the trial-type prediction method was replaced by the simple spatial reconstruction algorithm. Similarly, trial types predicted using this method matched actual trial types moderately well only in replays at the junction (Figure 7—figure supplement 2). To further check whether the predicted and actual trial types were coincidently matched, the probability of trial-type prediction across all replays occurring at the junction was analyzed on a time-window basis. There was a significant effect of the probability of trial-type prediction for five trial types (NA_LR, DA_LR, VD_RL, NA_RL, and DA_RL) in episodic replays (ANOVA, p < 0.01; Figure 7—figure supplement 3). The predicted trial types showing the highest probability matched actual trial types for NA_LR, VD_RL, NA_RL, and DA_RL trials. Post hoc comparisons using the Tukey honest significant difference (HSD) test indicated that the predicted trial types with the highest probability were significantly different from most of the others, suggesting that the matches were not coincidental.

At the junction, only a single visual cue was present in the visually guided trials. To test whether the trial type encoded in these replays was therefore just a representation of the available sensory cues, I examined whether the difference in the number of visual cues showing (one or both) could be used to distinguish VD from the other subtasks encoded in replays at the junction. The number of visual cues showing did not account for VD representation, even at the junction (Figure 8A; χ² test, p = 0.20). The barrier that appeared at the center of the stem only in DA trials may also represent a sensory cue, which would distinguish NA from DA trials. However, the subtask predicted in the center of the stem did not match actual NA and DA trials (Figure 8B; χ² test, p = 0.26). This suggests that trial-type representation in the replays is more than just a representation of external cues, and that it is internally generated. I therefore conclude that the internally generated replay represents the actual trial type—not only the upcoming path but also the actual subtask to be rewarded—when the trial type can be identified, but otherwise it evenly represents all trial types thus far experienced on the reactivated path.

Figure 8

Download asset Open asset

My investigations into whether the subtask and path experienced during task performance are represented in replays of place cell activity sequences associated with SWRs during brief immobility periods, produced the following findings. Firstly, the replay can represent non-spatial information on subtask as well as spatial information concerning the path. Secondly, whereas the path in the replay is encoded in temporally compressed firing timings across place cells, the accompanying subtask is encoded in their temporally compressed firing rates. Thirdly, the replay is enhanced by spatial working memory demand. Finally, the replay can represent the actual trial type accurately when the trial type can be identified, but otherwise it evenly represents all trial types previously experienced along the reactivated path.

Place cell activity can encode not only spatial information about past, current, and future places (Frank et al., 2000; Wood et al., 2000; Ferbinteanu and Shapiro, 2003) but also non-spatial information including task demand (Markus et al., 1995; Anderson and Jeffery, 2003; Smith and Mizumori, 2006), odors and their match/non-match status (Otto and Eichenbaum, 1992; Sakurai, 1996; Wood et al., 1999), and elapsed time (Pastalkova et al., 2008; MacDonald et al., 2011). The sequential activation of place cells reflects distinct episodic-like memories (Pastalkova et al., 2008); in fact, reactivation of the firing rates of single neurons in the human hippocampus represents specific episodes during free recall (Gelbard-Sagiv et al., 2008). However, such episodic-like memory traces are present across time-scales similar to those during the actual experience, and are thus distinct from the rapid, temporally compressed replays reported here. Although numerous previous reports have concluded that place cell activity sequences are reactivated in a temporally compressed manner in the awake state (Foster and Wilson, 2006; O'Neill et al., 2006; Diba and Buzsáki, 2007; Singer and Frank, 2009; Gupta et al., 2010) as well as during slow-wave sleep (Wilson and McNaughton, 1994), they have shown only that replays encode familiar paths. Until now, the possibility that replays simultaneously represent both spatial and non-spatial information has been purely conjectural. It is well known that spatial information is encoded in temporally compressed firing timings across place cells in replays. I have now shown that replays also encode non-spatial information in temporally compressed firing rates, revealing that global and rate remapping mechanisms during running (Leutgeb et al., 2005; Takahashi, 2013) are preserved in the temporally compressed replays that occur during brief periods of immobility.

Since awake replay is more common in a novel environment (Cheng and Frank, 2008), a few previous studies (O'Neill et al., 2006) have speculated that the content of replays reflects the total of previous experience. Contradicting with this hypothesis, growing evidence suggests that replays are not a simple function of experience (Gupta et al., 2010), but rather that reward outcomes enhance the reactivation of experiences (Singer and Frank, 2009) and that replays can depict future paths to remembered goals (Pfeiffer and Foster, 2013). In addition, awake replays often begin at the animal's current location (Diba and Buzsáki, 2007). The relationship between replay content and experience therefore does not seem to be straightforward. Some previous reports suggest that awake replay is similar to vicarious trial-and-error (VTE) events (Hu and Amsel, 1995; Johnson and Redish, 2007), because it can represent possible future paths even in completely unfamiliar places (Gupta et al., 2010). Although VTE events are reported to occur outside of SWRs, a subsequent study suggested that awake replay associated with SWRs is seen frequently at the choice point in a similar task, where animals must make a memory-guided decision between two journeys (Karlsson and Frank, 2009). Consistent with activity observed in VTE events, these awake replays extend out to the left or right arms of the track, as would be expected if the animal were playing out possible upcoming choices. Frank et al. therefore proposed a unified interpretation in which replays provide information on possible upcoming paths to downstream brain structures such as the prefrontal cortex and nucleus accumbens, which can assess the value of different paths and make a decision about future actions (Carr et al., 2011).

In this study, I trained rats to perform a similar task, but path representation in these replays was directly linked to the animal's future path. This apparent disparity may reflect the proficiency of the task in previous studies. In contrast to previous VTE studies (Johnson and Redish, 2007; Karlsson and Frank, 2009; Gupta et al., 2010), well-trained rats in the present study could accurately predict a goal-directed path at the behavioral level even at the decision point. However, the task design did not allow prediction of the subtask until the animal had entered the decision point. This suggests that not only the path but also the subtask representation in the replay is similar to VTE events. By extending the unified interpretation to include non-spatial memory in the retrieved memory, the fact that replays in the present study reflected all previously experienced trial types (except at the decision point) suggests that replays prepare the brain for unforeseen changes, which would allow continuous switching between subtasks.

I also found that replays were enhanced by spatial working memory demand. Since the initiation of an awake replay is often related to external spatial input at the animal's location (Karlsson and Frank, 2009), this enhancement of replays suggests that the internal demand of the hippocampal-dependent spatial working memory is another factor. Considering that this is similar to a cued memory retrieval process, my findings strongly support the hypothesis that awake replay plays a key role in episodic-like memory retrieval, which contributes to memory-based navigational planning and decision making (Carr et al., 2011).

The sequential activation of place cell activity associated with SWRs can also represent paths where the animal has never experienced (Dragoi and Tonegawa, 2011; Dragoi and Tonegawa 2013a; 2013b; Ólafsdóttir et al., 2015). As a result, a fairly large proportion of events representing different and multiple experiences can spontaneously occur, even in naive animals. I speculate that the reason why paths and subtasks were encoded in temporally compressed firing timings and rates, respectively, in the replays may be to arrange such multiple preplayed episodes into a single replay event to easily recall and imagine an upcoming episode based on prior experience. In the present study, the rats were sequentially trained in the VD, NA, and DA subtasks. Together with the preplay mechanism, this training protocol might have helped to recruit a consistent path representation across the three subtasks. Until now, previous studies investigating the content of preplays have primarily focused on the geography and timeline of the animals' running. Non-spatial information in the preplay may provide further insight into memory encoding and retrieval.

This study has shown that non-spatial subtask information occurring along temporally ordered places, as experienced by the animal, could be predicted by the sequential reactivation of place cells occurring while the animal paused during engagement in the task. I therefore speculate that awake replay is linked to the typical abilities of episodic memory: mental time travel and foreseeing future situations. This provides novel insight into debates on animals' capacities to use the faculties provided by episodic memory (Allen and Fortin, 2013).

1. 1. Allen K
   2. Rawlins JN
   3. Bannerman DM
   4. Csicsvari J

   (2012) Hippocampal place cells can encode multiple trial-dependent features through rate remapping

   Journal of Neuroscience 32:14752–14766.

   https://doi.org/10.1523/JNEUROSCI.6175-11.2012

   • Google Scholar
2. 1. Allen TA
   2. Fortin NJ

   (2013) The evolution of episodic memory

   Proceedings of the National Academy of Sciences of USA 110:10379–10386.

   https://doi.org/10.1073/pnas.1301199110

   • Google Scholar
3. 1. Anderson MI
   2. Jeffery KJ

   (2003)

   Heterogeneous modulation of place cell firing by changes in context

   The Journal of Neuroscience 23:8827–8835.

   • Google Scholar
4. 1. Benjamini Y
   2. Fonio E
   3. Galili T
   4. Havkin GZ
   5. Golani I

   (2011) Colloquium paper: Quantifying the buildup in extent and complexity of free exploration in mice

   Proceedings of the National Academy of Sciences of USA 108:15580–15587.

   https://doi.org/10.1073/pnas.1014837108

   • Google Scholar
5. 1. Carr MF
   2. Jadhav SP
   3. Frank LM

   (2011) Hippocampal replay in the awake state: a potential substrate for memory consolidation and retrieval

   Nature Neuroscience 14:147–153.

   https://doi.org/10.1038/nn.2732

   • Google Scholar
6. 1. Cheng S
   2. Frank LM

   (2008) New experiences enhance coordinated neural activity in the hippocampus

   Neuron 57:303–313.

   https://doi.org/10.1016/j.neuron.2007.11.035

   • Google Scholar
7. 1. Davidson TJ
   2. Kloosterman F
   3. Wilson MA

   (2009) Hippocampal replay of extended experience

   Neuron 63:497–507.

   https://doi.org/10.1016/j.neuron.2009.07.027

   • Google Scholar
8. 1. Diba K
   2. Buzsáki G

   (2007) Forward and reverse hippocampal place-cell sequences during ripples

   Nature Neuroscience 10:1241–1242.

   https://doi.org/10.1038/nn1961

   • Google Scholar
9. 1. Dragoi G
   2. Tonegawa S

   (2011) Preplay of future place cell sequences by hippocampal cellular assemblies

   Nature 469:397–401.

   https://doi.org/10.1038/nature09633

   • Google Scholar
10. 1. Dragoi G
    2. Tonegawa S

    (2013a) Development of schemas revealed by prior experience and NMDA receptor knock-out

    eLife 2:e01326.

    https://doi.org/10.7554/eLife.01326

    • Google Scholar
11. 1. Dragoi G
    2. Tonegawa S

    (2013b) Distinct preplay of multiple novel spatial experiences in the rat

    Proceedings of the National Academy of Sciences of USA 110:9100–9105.

    https://doi.org/10.1073/pnas.1306031110

    • Google Scholar
12. 1. Ferbinteanu J
    2. Shapiro ML

    (2003) Prospective and retrospective memory coding in the hippocampus

    Neuron 40:1227–1239.

    https://doi.org/10.1016/S0896-6273(03)00752-9

    • Google Scholar
13. 1. Foster DJ
    2. Wilson MA

    (2006) Reverse replay of behavioural sequences in hippocampal place cells during the awake state

    Nature 440:680–683.

    https://doi.org/10.1038/nature04587

    • Google Scholar
14. 1. Frank LM
    2. Brown EN
    3. Wilson M

    (2000) Trajectory encoding in the hippocampus and entorhinal cortex

    Neuron 27:169–178.

    https://doi.org/10.1016/S0896-6273(00)00018-0

    • Google Scholar
15. 1. Gelbard-Sagiv H
    2. Mukamel R
    3. Harel M
    4. Malach R
    5. Fried I

    (2008) Internally generated reactivation of single neurons in human hippocampus during free recall

    Science 322:96–101.

    https://doi.org/10.1126/science

    • Google Scholar
16. 1. Gupta AS
    2. van der Meer MA
    3. Touretzky DS
    4. Redish AD

    (2010) Hippocampal replay is not a simple function of experience

    Neuron 65:695–705.

    https://doi.org/10.1016/j.neuron.2010.01.034

    • Google Scholar
17. 1. Hassabis D
    2. Kumaran D
    3. Vann SD
    4. Maguire EA

    (2007) Patients with hippocampal amnesia cannot imagine new experiences

    Proceedings of the National Academy of Sciences of USA 104:1726–1731.

    https://doi.org/10.1073/pnas.0610561104

    • Google Scholar
18. 1. Hu D
    2. Amsel A

    (1995) A simple test of the vicarious trial-and-error hypothesis of hippocampal function

    Proceedings of the National Academy of Sciences of USA 92:5506–5509.

    https://doi.org/10.1073/pnas.92.12.5506

    • Google Scholar
19. 1. Jadhav SP
    2. Kemere C
    3. German PW
    4. Frank LM

    (2012) Awake hippocampal sharp-wave ripples support spatial memory

    Science 336:1454–1458.

    https://doi.org/10.1126/science.1217230

    • Google Scholar
20. 1. Jezek K
    2. Henriksen EJ
    3. Treves A
    4. Moser EI
    5. Moser MB

    (2011) Theta-paced flickering between place-cell maps in the hippocampus

    Nature 478:246–249.

    https://doi.org/10.1038/nature10439

    • Google Scholar
21. 1. Johnson A
    2. Redish AD

    (2007) Neural ensembles in CA3 transiently encode paths forward of the animal at a decision point

    The Journal of neuroscience 27:12176–12189.

    https://doi.org/10.1523/JNEUROSCI.3761-07.2007

    • Google Scholar
22. 1. Karlsson MP
    2. Frank LM

    (2009) Awake replay of remote experiences in the hippocampus

    Nature Neuroscience 12:913–918.

    https://doi.org/10.1038/nn.2344

    • Google Scholar
23. 1. Leutgeb S
    2. Leutgeb JK
    3. Barnes CA
    4. Moser EI
    5. McNaughton BL
    6. Moser MB

    (2005) Independent codes for spatial and episodic memory in hippocampal neuronal ensembles

    Science 309:619–623.

    https://doi.org/10.1126/science.1114037

    • Google Scholar
24. 1. MacDonald CJ
    2. Lepage KQ
    3. Eden UT
    4. Eichenbaum H

    (2011) Hippocampal “time cells” bridge the gap in memory for discontiguous events

    Neuron 71:737–749.

    https://doi.org/10.1016/j.neuron.2011.07.012

    • Google Scholar
25. 1. Markus EJ
    2. Qin YL
    3. Leonard B
    4. Skaggs WE
    5. McNaughton BL
    6. Barnes CA

    (1995)

    Interactions between location and task affect the spatial and directional firing of hippocampal neurons

    The Journal of Neuroscience 15:7079–7094.

    • Google Scholar
26. 1. McNaughton BL
    2. Barnes CA
    3. O’keefe J

    (1983) The contributions of position, direction, and velocity to single unit activity in the hippocampus of freely-moving rats

    Experimental Brain Research 52:41–49.

    https://doi.org/10.1007/BF00237147

    • Google Scholar
27. 1. O'Keefe J
    2. Dostrovsky J

    (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat

    Brain Research 34:171–175.

    https://doi.org/10.1016/0006-8993(71)90358-1

    • Google Scholar
28. 1. O'Neill J
    2. Senior T
    3. Csicsvari J

    (2006) Place-selective firing of CA1 pyramidal cells during sharp wave/ripple network patterns in exploratory behavior

    Neuron 49:143–155.

    https://doi.org/10.1016/j.neuron.2005.10.037

    • Google Scholar
29. 1. Ólafsdóttir HF
    2. Barry C
    3. Saleem AB
    4. Hassabis D
    5. Spiers HJ

    (2015) Hippocampal place cells construct reward related sequences through unexplored space

    eLife 4:e06063.

    https://doi.org/10.7554/eLife.06063

    • Google Scholar
30. 1. Otto T
    2. Eichenbaum H

    (1992) Neuronal activity in the hippocampus during delayed non-match to sample performance in rats: evidence for hippocampal processing in recognition memory

    Hippocampus 2:323–334.

    https://doi.org/10.1002/hipo.450020310

    • Google Scholar
31. 1. Pastalkova E
    2. Itskov V
    3. Amarasingham A
    4. Buzsáki G

    (2008) Internally generated cell assembly sequences in the rat hippocampus

    Science 321:1322–1327.

    https://doi.org/10.1126/science.1159775

    • Google Scholar
32. 1. Pfeiffer BE
    2. Foster DJ

    (2013) Hippocampal place-cell sequences depict future paths to remembered goals

    Nature 497:74–79.

    https://doi.org/10.1038/nature12112

    • Google Scholar
33. 1. R Development Core Team

    (2013) R: A language and environment for statistical computing

    R: A language and environment for statistical computing, Available at: http://www.R-project.org/.

    http://www.R-project.org/

    • Google Scholar
34. 1. Sakurai Y

    (1996)

    Hippocampal and neocortical cell assemblies encode memory processes for different types of stimuli in the rat

    The Journal of Neuroscience 16:2809–2819.

    • Google Scholar
35. 1. Singer AC
    2. Frank LM

    (2009) Rewarded outcomes enhance reactivation of experience in the hippocampus

    Neuron 64:910–921.

    https://doi.org/10.1016/j.neuron.2009.11.016

    • Google Scholar
36. 1. Smith DM
    2. Mizumori SJ

    (2006) Learning-related development of context-specific neuronal responses to places and events: the hippocampal role in context processing

    The Journal of Neuroscience 26:3154–3163.

    https://doi.org/10.1523/JNEUROSCI.3234-05.2006

    • Google Scholar
37. 1. Takahashi S

    (2013) Hierarchical organization of context in the hippocampal episodic code

    eLife 2:e00321.

    https://doi.org/10.7554/eLife.00321

    • Google Scholar
38. 1. Takahashi S
    2. Anzai Y
    3. Sakurai Y

    (2003) Automatic sorting for multi-neuronal activity recorded with tetrodes in the presence of overlapping spikes

    Journal of Neurophysiology 89:2245–2258.

    https://doi.org/10.1152/jn.00827.2002

    • Google Scholar
39. 1. Takahashi S
    2. Sakurai Y

    (2005) Real-time and automatic sorting of multi-neuronal activity for sub-millisecond interactions in vivo

    Neuroscience 134:301–315.

    https://doi.org/10.1016/j.neuroscience.2005.03.031

    • Google Scholar
40. 1. Takahashi S
    2. Sakurai Y

    (2009) Sub-millisecond firing synchrony of closely neighboring pyramidal neurons in hippocampal CA1 of rats during delayed non-matching to sample task

    Frontiers in Neural Circuits 3:9.

    https://doi.org/10.3389/neuro.04.009.2009

    • Google Scholar
41. Book

    1. Tulving E

    (1983)

    Elements of episodic memory

    New York: Oxford University Press.

    • Google Scholar
42. 1. Wiener SI
    2. Paul CA
    3. Eichenbaum H

    (1989)

    Spatial and behavioral correlates of hippocampal neuronal activity

    The Journal of Neuroscience 9:2737–2763.

    • Google Scholar
43. 1. Wilson MA
    2. McNaughton BL

    (1994) Reactivation of hippocampal ensemble memories during sleep

    Science 265:676–679.

    https://doi.org/10.1126/science.8036517

    • Google Scholar
44. 1. Wood ER
    2. Dudchenko PA
    3. Eichenbaum H

    (1999) The global record of memory in hippocampal neuronal activity

    Nature 397:613–616.

    https://doi.org/10.1038/17605

    • Google Scholar
45. 1. Wood ER
    2. Dudchenko PA
    3. Robitsek RJ
    4. Eichenbaum H

    (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location

    Neuron 27:623–633.

    https://doi.org/10.1016/S0896-6273(00)00071-4

    • Google Scholar
46. 1. Zhang K
    2. Ginzburg I
    3. McNaughton BL
    4. Sejnowski TJ

    (1998)

    Interpreting neuronal population activity by reconstruction: unified framework with application to hippocampal place cells

    Journal of Neurophysiology 79:1017–1044.

    • Google Scholar

Author details

1. Susumu Takahashi

   Laboratory of neural circuitry, Graduate School of Brain Science, Doshisha University, Kyoto, Japan

   Contribution

   ST, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   stakahas@mail.doshisha.ac.jp

   Competing interests

   The author declares that no competing interests exist.

• 3,104

  views

• 522

  downloads

• 31

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.